Residual bootstraps for regression model validation

Abstract

Validation is a useful and necessary part of the model-building process, identification of one or several “good” regression models is not the end of the model-building process and these models must be evaluated by various diagnostic procedures before the final regression model is determined. Residual Bootstrap method in regression model validation accomplish the goal of constructing appropriate sampling distributions empirically using the data at hand instead of statistician relying on theoretical sampling distributions like the normal, t and f where appropriateness for any given problem always rest on untestable assumptions. Validation statistics of interest such as standard error (SE), mean square error (MSE) and coefficient of determination (?²) were used as criteria for selecting the best model suitable for predictive purposes.The research work concluded that to reduce the problem of overffited models in regression analysis, residual bootstrap approach should be employed in checking the validation of regression model as it gives a better estimates and stable value of coefficient of determination.